Kimi K3 scores 32% on ExploitBench, far behind U.S. frontier models
The Decoder · rss · 2026-07-24
A report from The Decoder says Moonshot AI's Kimi K3 performed far worse than frontier U.S. models on offensive cyber tasks.
According to the British AI Security Institute and the U.S. Center for AI Standards and Innovation:
- 32% on ExploitBench for Kimi K3
- 76% for leading U.S. models
- safeguards did not reliably block exploit development or simulated attacks
The article argues the gap between Kimi K3's strong general benchmarks and weak cyber performance may align with allegations that Moonshot AI distilled Anthropic models.
More from Models
- Rumor claims Kimi faked performance by serving Claude; DeepSeek new model surprises in evals — realsohamparekh · 2026-09-11
- GPT-5.6 writes well but is instantly forgettable, user complains — BasedRaddka · 2026-09-11
- Opus Refuses Protein Research Codebase Over 'Safety' Concerns, Dev Considers Rolling His Own — josephdviviano · 2026-09-11
- User Hails Unconfirmed 'DeepSeek 4.1 Flash' as an Inflection Point in LLMs — himanshustwts · 2026-09-11
- Terminal Bench v4: GLM-5.3 Leads at 41.9%, Kimi-K3 Underwhelms at 12.6% — Ok_Warning2146 · 2026-09-11
- GPT-6 Astra beats Factorio with enemies in 44 in-game hours at ~$4,500 API cost — liminal_bardo · 2026-09-11